Machine Learning Using Driver Monitoring Cameras to Detect Drunk Driving

Drunk driving is a major cause of road traffic injuries and fatalities. To address this, we support recent efforts for driver monitoring systems, which are deemed a promising means to detect impaired driver states. In this paper, we develop an in-vehicle machine learning system to determine the blood alcohol concentration (BAC) levels of drivers. Specifically, our system leverages driver monitoring cameras, which are nowadays mandated in numerous countries worldwide, to predict critical BAC thresholds. We evaluate our system with n=30 participants in an interventional clinical study with a research-grade driving simulator. Our system detects driving under moderate levels of alcohol influence (i.e., 0.00 g/dL < BAC ≤ 0.03 g/dL) with an area under the receiver operating characteristic curve (AUROC) of 0.88 and driving above the WHO recommended legal BAC threshold of 0.05 g/dL with an AUROC of 0.79. By inspecting the trained machine learning model, we find that our system relies on known pathophysiological effects associated with alcohol consumption. Our results highlight the potential of driver monitoring cameras to determine BAC levels and then to intervene to prevent accidents caused by drunk driving.